For many businesses, AI adoption has happened one tool at a time. Employees have gained new ways to automate or accelerate individual tasks, but those tools often sit within the same processes, systems, and organisational boundaries that were already in place.
The challenge now is to bring more structure to how AI is deployed across the business. Data from Gartner, based on a survey of 1,300 leaders, found that just 22% of enterprises have successfully scaled AI across multiple business departments. Meanwhile, just 37% of 1,700 leaders surveyed by McKinsey said AI had increased profitability.
Fragmentation also creates governance and security challenges. Leaders need human oversight and clear controls over AI access to sensitive business and customer data.
Leaders must move beyond AI tools that shave time off individual tasks and use the technology to improve workflows across the business. That creates opportunities for greater efficiency and profitability.
That shift matters as enterprise AI evolves from language models that help employees complete individual tasks to agents that can act across multiple systems and stages of work. Their value depends on how effectively those actions connect into an end-to-end workflow built around a clear business outcome.
Why businesses need intelligent workflows
OneAdvanced, a provider of AI-powered, sector-focused SaaS software, is helping organisations navigate that challenge.
Marko Perisic, the firm’s Chief Product Officer, says the current use of AI fails to create the intelligent, enterprise-wide change needed to drive tangible outcomes.
“Most organisations, unfortunately, go down the path of getting an AI subscription of some sort, giving it to employees, and assuming that it will improve how they work and how they produce output,” he says. “But when you just apply AI on top of archaic workflows and sub-optimal ways of working, all you get is an AI supercharged mess. In a lot of cases, this creates more problems and compliance issues than positive outcomes.”
Organisations seeing the greatest benefits from AI are using it to rethink existing workflows or design new ones that connect work across the business.
Perisic says doing this requires a deliberate approach. “You need to go down a path of conscious, purposeful, and structured transformation,” he says. “You need to rethink those workflows from the ground up and ask: ‘What do we actually want to try to accomplish?’ In the world of AI, perhaps you don’t need Step One, Three, and Seven of that workflow. Maybe you only need Two and Four, but they need to be very different from what they used to be. When you do that, that’s when you get the real benefits of AI.”
Designing persona-first workflows
OneAdvanced works with organisations to create persona-first workflows. This means designing the way work gets done around the specific needs, responsibilities, and context of the person using the software. The aim is to make technology fit the way people work, instead of forcing employees to adapt their work to a proliferation of AI tools.
The process begins by identifying a specific role, such as a Financial Controller or Chief Marketing Officer, and mapping the key workflows involved. OneAdvanced then assesses the potential return in time, efficiency, or cost, and prioritises those with the greatest value for the customer.
“It’s not enough to just analyse one step of a workflow,” says Perisic. “You need to really understand what can be done better using AI, what can be done faster using AI, how to do it in a safe and compliant way, how to do it in a way where it actually fits with the rest of the organisational workflows that maybe aren’t AI-powered at the same time, and where the interface points.”
An intelligent workflow in action
OneAdvanced IQ is designed to put this approach into practice, connecting people, data, workflows, and AI in one system of work.
A law firm recently partnered with OneAdvanced to use IQ to automate client onboarding. The process was time-consuming and non-billable, but essential for compliance, including anti-money laundering checks. Automating parts of the workflow allowed legal experts to focus on higher-value work.
In practice, that means connecting the different stages of onboarding rather than automating a single check in isolation. Once a new-client request triggers the workflow, AI agents can gather the required information, carry out the relevant checks, flag exceptions, and return the evidence for review. Where a result is ambiguous or requires professional judgement, the agent will escalate it to the appropriate person rather than making the decision autonomously.
“It’s had a huge transformational impact for these individuals,” says Perisic. “They can tell the AI about the client they want to onboard and the checks they need to make, and then AI agents will run that workflow in the background. At the end, it will report back with a list of completed checks, and then the client can be successfully onboarded. Legal professionals can then do the work they’re trained for, minimising administrative tasks.”
But while AI can be used to carry out workflows autonomously, human oversight is still critical to ensure the security and governance of sensitive customer data. Perisic says organisations should begin by limiting the scope of AI. “You should start super narrow with very limited permission sets,” he explains. “Then you can consciously expand that with controls, so that people understand that agents have access to certain data sets or have permission to send emails or carry out other tasks.”
Permissions form one part of a wider control framework. Organisations need visibility into where sensitive data is held and processed. They should also be able to understand and explain how AI reaches decisions or recommendations, with clear checkpoints for human review, approval, or intervention. Safeguards should reflect the potential impact on customers, patients, or citizens.
Organisations can start with a tightly defined, high-value workflow, establish how it performs today, and automate specific stages while retaining human controls. They can then measure the impact, identify where exceptions arise, and gradually expand the agent’s remit. This limits disruption and gives employees time to adapt.
The next phase of enterprise AI will depend less on how many tools organisations deploy and more on how effectively they connect AI to the work that people already do.
Five questions on AI agents
Marko Perisic, Chief Product Officer at OneAdvanced, explains how leaders can introduce AI agents with clear goals, controls and human oversight.
Before deciding which AI model or fleet of agents to use, what business outcomes should leaders define?
MP: Ultimately, every business wants to increase profitability, but that’s a lagging indicator. You need to work backwards from there to identify the leading indicators of success.
For example, if you want to increase profitability, do you want to do that by reducing costs? Or do you want to do that by increasing growth? That will be one decision you need to make. If it’s increasing growth, how do you increase growth? You need to increase your win rate. How do you increase your win rate? You do so by competing, offering better products and services in that particular target market segment. And you work yourself backwards.
Eventually, you’ll arrive at the leading indicators of success. And that could be increased usage, better NPS from customers, better customer sentiment that eventually is going to lead down the chain to positive economic indicators of performance of the business.
How can businesses measure whether AI is actually improving a workflow versus just improving the ability to perform individual tasks?
MP: It comes back to lagging and leading indicators. If AI is not changing the ultimate measures of a successful business (profitability) either by reducing costs or increasing growth, then it’s not really useful.
Now, it may reduce the time that employees need to spend doing work or learning how to do work. So that will increase employee satisfaction, which in turn can also benefit the economic performance of the business. You get better employee retention. You get better engagement, and better-engaged employees produce better work.
Ultimately, it really comes down to growth and profitability. So, I think you just need to keep that in mind when you look at it. Because otherwise, as I said, you’re just burning tokens and everybody gets busy, but you’re not producing the output you want to produce.
What key boundaries should businesses set if they’re using agents for the first time?
MP: First of all, I’d make sure you have a very purposeful and thoughtfully designed security posture in the organisation and a dedicated team that’s managing that.
You need to understand where your critical assets are, where your critical data is, who has permission to access it, and how and when they access it. You need to narrow those permissions down as much as possible.
One approach is to adopt zero-based security. This is basically no permissions, with permissions only granted by exception and acceptance. You need to start narrow and expand rather than give everyone access and mitigate when it goes wrong. Because when it goes wrong, it’s already too late. And it’s harder to untangle.
Where should human judgment and oversight remain?
MP: I think it needs to be measured by AI’s impact on other human lives. All businesses have an impact on human lives. If my pizza delivery is half an hour late because some agent screwed up, it’s not the worst thing in the world. But if my medical prescription is wrong because an agent prescribed the wrong medicine based on a diagnosis it produced, then that could be fatal. And that needs to be very critically controlled and addressed.
The severity of the potential impact on people’s lives should determine the level of human control. And also, for that very reason, there are laws and compliance rules around some of these things. What can be used in certain industries and what cannot. We adhere to those very strictly and thoroughly. I think there’s going to be more and more focus on that in the industry because there’s a lot of AI sprawl. If it’s not controlled, it could be damaging.
In the first 90 days, what should I measure to understand whether AI implementation has been a success?
MP: Be clear about your lagging indicators and the leading indicators you’re going to measure. Consider the severity of the potential impact on people, how many controls the workflow needs, and who owns them.
Document this in a structured, centralised place that the right people can access. That gives you a foundation for deciding how to move forward. Giving employees AI, hoping for the best, and mitigating problems afterwards isn’t the way to go. You need to start with a structured view.
For many businesses, AI adoption has happened one tool at a time. Employees have gained new ways to automate or accelerate individual tasks, but those tools often sit within the same processes, systems, and organisational boundaries that were already in place.
The challenge now is to bring more structure to how AI is deployed across the business. Data from Gartner, based on a survey of 1,300 leaders, found that just 22% of enterprises have successfully scaled AI across multiple business departments. Meanwhile, just 37% of 1,700 leaders surveyed by McKinsey said AI had increased profitability.
Fragmentation also creates governance and security challenges. Leaders need human oversight and clear controls over AI access to sensitive business and customer data.
